Show error output on tool failure and print final result to stdout

src/council/artifacts.py

@@ -75,15 +75,17 @@ def _source_to_dict(src: ContextSource) -> dict:
 
 
 def save_round0(run_dir: Path, prompts: dict[str, str], results: dict[str, ToolResult]) -> None:
-    """Save Round 0 artifacts (parallel generation)."""
+    """Save Round 0 artifacts (parallel generation).
+
+    Saves all tool results by their actual names (supports multi-candidate
+    names like ``claude_0``, ``codex_1`` in addition to plain ``claude``).
+    """
     rdir = run_dir / "rounds" / "0_generate"
-    for name in ("claude", "codex"):
-        if name in prompts:
-            (rdir / f"prompt_{name}.md").write_text(prompts[name], encoding="utf-8")
-        if name in results:
-            r = results[name]
-            (rdir / f"{name}_stdout.md").write_text(r.stdout, encoding="utf-8")
-            (rdir / f"{name}_stderr.txt").write_text(r.stderr, encoding="utf-8")
+    for name, prompt_text in prompts.items():
+        (rdir / f"prompt_{name}.md").write_text(prompt_text, encoding="utf-8")
+    for name, r in results.items():
+        (rdir / f"{name}_stdout.md").write_text(r.stdout, encoding="utf-8")
+        (rdir / f"{name}_stderr.txt").write_text(r.stderr, encoding="utf-8")
 
 
 def save_round(

src/council/pipeline.py

@@ -256,6 +256,9 @@ async def run_pipeline(opts: RunOptions, config: CouncilConfig) -> Path:
     if patch:
         _print_progress(f"  Patch file:   {run_dir / 'final' / 'final.patch'}")
 
+    # Print final output to stdout so the user sees the answer directly.
+    print(final_output)
+
     return run_dir
 
 
@@ -364,6 +367,8 @@ async def resume_pipeline(
     if patch:
         _print_progress(f"  Patch file:   {run_dir / 'final' / 'final.patch'}")
 
+    print(final_output)
+
     return run_dir
 
 
@@ -457,10 +462,16 @@ def _should_run(round_name: str) -> bool:
         for name, result in r0_results.items():
             _print_progress(f"  {name}: {_tool_status_str(result)} ({result.duration_sec:.1f}s)")
             _print_verbose(f"stdout: {len(result.stdout)} bytes, stderr: {len(result.stderr)} bytes", verbose)
-            # Show stderr when a tool fails so users can diagnose issues.
-            if not _tool_ok(result) and result.stderr.strip():
-                for line in result.stderr.strip().splitlines()[:10]:
-                    _print_progress(f"    [dim]{line}[/dim]")
+            # Show error output when a tool fails so users can diagnose issues.
+            # Check both stderr and stdout since some tools (e.g. Claude Code)
+            # write errors to stdout in print mode.
+            if not _tool_ok(result):
+                for stream, label in ((result.stderr, "stderr"), (result.stdout, "stdout")):
+                    text = stream.strip()
+                    if text:
+                        _print_progress(f"    [dim]{label}:[/dim]")
+                        for line in text.splitlines()[:10]:
+                            _print_progress(f"    [dim]  {line}[/dim]")
 
         # Collect successful outputs grouped by tool family.
         claude_candidates: list[tuple[str, str]] = []